ECP 2022 Abstract Book

Virchows Archiv (2022) 481 (Suppl 1):S1–S364 13 learning model was able to learn from the masks and found the fibrotic areas in the HE stained digital sections with an associated F-score of 0.76 in the validation data set. Conclusion: The proposed method is feasible and can provide a rapid and reproducible quantitative result for interstitial fibrosis in HE stained kidney biopsies. We will train and fine-tune the deep learning model with more data and expect to see even better performance. The model will then be tested for robustness in an independent cohort. Funding: Western Norway Regional Health Authority OFP-05-015 Relationship between immunosuppressive treatment, morphol- ogy, and gene expression in T-cell-mediated rejection of the transplanted kidney D. Dobi*, S. Chandran, J.R. Greenland, F. Vincenti, Z. Laszik *Semmelweis University, Hungary Background & objectives: Belatacept preserves renal function better than calcineurin inhibitor (CNI)-based immunosuppression. However, higher frequency of first-year T-cell-mediated rejection (TCMR) in belatacept-treated patients hampered the adoption of costimulation blockade. We set out to study the patomechanism of TCMR in this patient group. Methods: Formalin-fixed paraffin-embedded renal biopsy samples were analysed from 92 patients stratified by histopathologic diagnosis (TCMR, borderline changes, or normal) and immunosuppression regimen (belatacept, CNI). We applied gene expression analysis and whole slide inflammatory cell quantification to assess the impact of belatacept on intragraft immune signature. Results: Ninety-one percent of genes overexpressed in TCMR showed significant correlation with whole-section inflammatory load. There were 27 genes that had a positive association with belatacept treatment. These were mostly related to myeloid cells and innate immunity. Genes negatively associated with costimula- tion blockade (n=14) could be linked to B-cell differentiation and proliferation. Conclusion: We concluded that expression levels of genes char- acteristic of TCMR are strongly interconnected with quantitative changes of the biopsy inflammatory load. Our results might sug- gest differential involvement of the innate immune system, and an altered B-cell engagement during TCMR in belatacept-treated patients relative to CNI-treated referents. OFP-06 | Joint Oral Free Paper Session Pulmonary Pathology / Thymic and Mediastinal Pathology OFP-06-001 Multi-case learning model for predicting EGFR and KRAS gene mutation in non-small cell lung cancer Y. Ding, C. Wu, Y. Zhao, J. Yao, Y. Liu* *The Fourth Hospital, China Background & objectives: To develop an artificial intelligence learning model for predicting EGFR and KRAS gene mutation in non-small cell lung cancer (NSCLC) by integrating the information of pathological images. Methods: 934 NSCLC biopsy whole slice images (WSIs) were collected. EGFR and KRAS gene mutation were detected by next- generation sequencing (NGS). The WSIs were divided into train- ing set, validation set and test set, and a transformer-based multi- instance learning (T-MIL) model was developed to predict EGFR and KRAS gene mutation. Moreover, T-MIL model was compared with the other models. Results: The area under the cure (AUC) was 0.711 by using T-MIL model to predict EGFR gene mutation, and the sensitivity, specific- ity, the positive predictive value (PPV), and the negative predictive value (NPV) were 81.6%, 55.6%, 61.7%, 77.5%, respectively. For predicting KRAS gene mutation, T-MIL model AUC value was 0.601, and the sensitivity, specificity, PPV, and NPV were 56.2%, 65.1%, 13.2%, 94.0%, respectively. Compared with other learn- ing models, including attention-based multiple instance learning (A-MIL) and RNN architecture for bag representation generation in MIL (RNN-MIL), T-MIL model demonstrated better performance. For EGFR gene mutation, the AUC value were 0.485, 0.6767, 0.711 respectively, and 0.5753, 0.5593, 0.601 respectively for KRAS gene mutation. Conclusion: We developed a T-MIL learning model for predicting NSCLC gene mutation, and demonstrated well performance. Its performance in predicting EGFR mutation was better than KRAS(AUC 0.711 vs 0.601). Our research proved that the performance of T-MIL learning model through pathological images was better than A-MIL and RNN-MIL model in predicting key driver gene mutation. OFP-06-002 Three cancer associated fibroblasts subtypes are associated with histological features, immune environment and prognosis in resected non-small cell lung cancer (NSCLC) E. Guenzi*, K. El Husseini, A. Guyard, A. Mailleux, P. Mordant, A. Couvelard, B. Crestani, A. Cazes, G. Zalcman, N. Pote *Department of Pathology, Hôpital Bichat, Assistance Publique - Hôpitaux de Paris, France Background & objectives: Three cancer-associated fibroblasts (CAFs) subtypes have been recently identified in breast cancer, characterized by the differential expression of FAP and ANTXR1. We aimed to assess, in NSCLC, the association of these CAF subtypes with histological features, immune environment and prognosis. Methods: Expression of FAP and ANTXR1 was assessed by immunohistochemistry (H-score) on tissue micro-array built from a retrospective series of 186 NSCLC surgical samples. Three CAF subtypes were defined by the differential expression of FAP and ANTXR1 (FAPLow; FAPHigh/ANTXR1Low; FAPHigh/ ANTXR1High) and correlated with histological features, immune environment (assessed by CD8, CD4, and FOXP3 expression) and prognosis. Results: 82 adenocarcinomas (ADC) and 104 squamous cell carcinomas (SCC) were included. We found a predominance of FAPHigh/ANTXR1High CAFs in SCC and FAPLow CAFs in ADC (p<0.001). In SCC, FAPhigh/ANTXR1low CAFs were associated with a higher CD8/CD4+CD8 ratio (p=0.02), but there was no correlation of CAFs subtypes and the immune envi- ronment in ADC. In ADC, a higher proportion of FAPHigh/ ANTXR1High CAFs was detected in poorly differentiated tumours (p<0.001). Finally, in ADC, tumours with FAPHigh/ ANTXR1High CAFs were associated with a poorer disease-free survival in patients that did not have adjuvant chemotherapy (p<0.05). In SCC, no association of CAFs subtypes with disease free survival was found. Conclusion: These three CAF subtypes are differentially expressed between ADC and SCC, and are associated with the immune envi- ronment in SCC and with tumour differentiation and disease free survival in ADC. These preliminary results suggest that FAP and ANTXR1 could be used as prognostic biomarkers in ADC. S23